Skip to main content
Fraport logo

CUSTOMER STORY

Fraport Scales Geospatial Radar Analytics on Databricks

Airport Analytics Ahead of Schedule

2 months ahead of schedule

Migrated geospatial radar analytics off a custom PostgreSQL database in-house, using AI coding tools to beat a one-year plan

Up to 2x faster reports

Moved Power BI reporting onto Databricks Lakehouse warehouses, speeding up reports on larger tables and meeting a 3-second target

100+ daily pipelines

Runs about 100 scheduled pipelines a day, plus roughly 1,000 external jobs, on one lakehouse governed by Unity Catalog

Dedicated compute for every analytics use case really helps us to have a stable platform.

- Martin Herold, Product Manager, Business Intelligence Platform, Fraport AG

Fraport AG operates Frankfurt Airport and manages airports worldwide. Its BI platform team needed to move a compute-heavy geospatial workload- the daily analysis of radar data that tracks aircraft on the apron- off a custom PostgreSQL database where it blocked other users. By migrating to the Databricks Lakehouse, Fraport gave each analytics use case dedicated compute and is on track to complete the one-year migration two months ahead of schedule.

Heavy geospatial jobs were blocking a shared database

Fraport has spent recent years modernizing its IT infrastructure and moving workloads to the cloud. Data volumes keep rising, driven in large part by sensor and IoT data, and more teams across the company want access to that data sooner. "Getting more data, getting the data quicker, getting more people using the data. These are the challenges on my desk," said Martin Herold, Product Manager for the Business Intelligence Platform at Fraport AG.

One of the most demanding workloads is geospatial analysis of radar data that shows how aircraft move on the apron. Fraport uses this analysis to optimize tracks and plan maintenance activities. The work ran on a custom database built on PostgreSQL with PostGIS extensions. Its daily jobs took about two hours and pushed the database to 100% CPU load, blocking other users who depended on the same system.

The team tried load balancing and resource limits, but those measures slowed the jobs down again. The workload needed its own compute. Running separate solutions also added maintenance effort, and the team wanted to scale compute and licenses up for a proof of concept and back down without carrying fixed costs. "Today I need GPUs and tomorrow maybe not. The change of the workloads is so dynamic that it makes sense to be on a hyperscaler," Herold said.

Dedicated compute for every workload on one governed lakehouse

Databricks was already at the center of Fraport's data estate. A data lake governed by Unity Catalog feeds Power BI reporting through Databricks Lakehouse warehouses, about 30 custom-built applications, RStudio users and SAS analysts. Moving geospatial processing onto the same platform gave each analytics use case dedicated compute and reduced the number of systems to operate. "Bringing more use cases in and getting rid of old things makes it easier for the operations team," Herold said.

The team decided that processing the data on the lakehouse with the Spark engine would scale better and keep complexity low. With Databricks' high-performance Spatial SQL functionality, they could start with standard capabilities rather than fully custom code. Fraport ran the migration with its own staff and no external partner, drawing on Databricks for concept discussions. For vendor applications, such as airport simulation software that only supports PostgreSQL connections, Fraport stages results in Lakebase with the PostGIS extension, so those consumers simply receive a new connection URL.

Git integration and collaborative workflows let data scientists from the business department work on the migration alongside IT. "The collaborative workflows helped us a lot to do the thing together and have a bigger workforce on it," Herold said. Engineering is done as code in notebooks, with jobs and pipelines defined in Terraform. Serverless compute is the default, with classic compute reserved for internal sources behind network restrictions. Developers used GitHub Copilot and Genie to convert code and write notebooks.

Faster, more flexible analytics ahead of schedule

Daily geospatial jobs now run on dedicated compute without blocking reporting or other workloads. The team can also tune each job for speed or cost, for example by giving a job more time over a weekend on less expensive compute. "We get new ways to get the same results even faster or even cheaper, which is something that you can't decide once for a year. Maybe something you want to decide every day," Herold said.

Reporting has improved as well. Herold estimates that Power BI reports are generally 1.5 to 2 times faster than on the previous database, with the largest gains on bigger tables, and reports meet the three-second open target most of the time. The platform runs about 100 scheduled pipelines each day, plus roughly 1,000 daily jobs from external ETL tools and applications that use the data lake.

The migration had a one-year target. By October, QA systems were complete and the QA database had been shut down, leaving only production and putting the project about two months ahead of schedule. Herold credits AI-assisted development and his team, which adopted AI fast, for much of that pace. "The benefit of AI helping developers was the best help we could have this year," he said.

Business users are adopting Genie agents to query data directly, and Herold sees that usage growing fastest across the platform. He summarized the value simply: "It's the platform that helps us to solve our current data issues and also the future ones."

Explore more

Frequently asked questions about Lakehouse

Fraport AG, the operator of Frankfurt Airport, moved its daily geospatial analysis of apron radar data from a custom PostgreSQL database with PostGIS extensions to the Databricks Lakehouse. Fraport's own staff ran the migration with no external partner, using the Spark engine and Databricks Spatial SQL functions instead of fully custom code. Developers used GitHub Copilot and Genie to convert code and write notebooks, putting the one-year migration about two months ahead of schedule.

Want to learn more about Lakehouse?